Channel lifecycle
A buffered channel accepts items until it is completed, finishes the items it accepted, and is then done. These are the states, and what ends an item.
- Running.
TryWriteandWaitToWriteAsyncaccept items. - Completing. The inbound channel is closed. The items that were accepted are still moving through the inbound buffer, the outbound channel and the exports.
- Drained. Every accepted item reached an end.
WaitForDrainAsyncreturnstrue. - Disposed. Terminal. A waiter for the drain is released with
false, unless the channel had drained.
The transitions are compare-and-swap on one value, so two threads cannot both move the channel on.
An item is counted when it is accepted, and stops being counted at exactly one of these ends:
| End | What the caller sees |
|---|---|
| Exported | The export returned a response in which the item did not ask to be retried. |
| Exhausted | The retry budget ran out, or the export threw and was not retried. The item is reported through ExportMaxRetriesCallback. |
| Dropped by the inbound channel | BufferItemDropped, when the full mode drops items. |
| Rejected by the server | ServerRejectionCallback. The item was exported, with a status that is not retryable. |
The channel is drained when the count reaches zero after completion. Nothing estimates how many exports are left, and nothing polls.
An exception that escapes ExportAsync is not a response the server gave. The transport retries the failures it models itself and lets an unexpected exception bubble out, so by default the items are given up on and reported, and not sent again with unknown idempotence. Set BufferOptions.ExportRetryOnException to say which exceptions are safe to retry for your data.
IncrementalSyncOrchestrator and DeltaSyncOrchestrator are started once.
A second StartAsync throws. After a failed start there is nothing to write to or complete, and the orchestrator says so. CompleteAsync returns false, without deleting documents or moving aliases, when a channel did not drain in time.